Claude Code On-the-Go

Original Article ↗ Hacker News Discussion ↗

On-the-go Claude / tooling setups

  • Many commenters describe similar “Claude-on-the-go” setups using Tailscale + SSH (Termius/Blink/WebSSH) + tmux to keep Claude Code sessions alive and reachable from phones or tablets.
  • Alternatives include Claude Code for Web via the Claude app, GitHub-based agents (Claude in GitHub, Copilot Agent, Jules), and hosted environments like Happy, Hapi, exe.dev, Opencode, Catnip, and others.
  • Some run everything on home machines over Tailscale to avoid VPS costs; others prefer disposable/cloud VMs for isolation and always-on access.

Planning, multi-agent, and workflow design

  • Debate over Claude Code Web’s missing “planning mode”; some emulate it by having Claude write spec.md or use native plan files under ~/.claude/plans.
  • Multi-feature work is handled via git worktrees, tmux windows, conductor-like orchestration tools, kanban-style boards, or multi-agent frameworks (e.g., PAL, Ralph-Wiggum, clawdbot).
  • Several people run multiple agents in parallel per repo/feature and mainly curate memory files and review PRs.

Code quality, testing, and review

  • Some are comfortable letting agents run “for hours” and only reviewing PR diffs at the end, often with automated preview deployments or CI.
  • Others insist that serious work still requires local checkouts, manual testing, and deep interaction with running services; they worry that web sandboxes without local code visibility feel like “driving blind” and encourage “slop.”
  • There’s concern about how to handle large diffs, subtle regressions, or complex products from a phone-sized UI.

Security, infra, and cost

  • Questions around SSH key forwarding into cloud VMs; suggestions to use scoped PATs, self-hosted runners, or proxy setups so containers never see real credentials.
  • Some argue the €7/day VM used in the post is unnecessarily expensive and advocate shorter-lived VMs, home hardware, or cheaper managed offerings.

Ergonomics: phones vs deep work

  • Split views: some find mobile + voice dictation (WisprFlow, Dictate, Parakeet, etc.) surprisingly effective for steering agents; others hate typing on phones and see mobile as only good for quick nudges and notifications.
  • Several say high-quality engineering still needs a big screen, keyboard, and focused time; mobile is best for “vibe coding” side projects or monitoring long-running tasks.

Labor, work-life balance, and hype

  • Strong anxiety that 24/7 agentic coding normalizes permanent availability and accelerates deterioration of white-collar work; countered by appeals to personal boundaries, labor organizing, or skepticism that LLMs are yet transformative.
  • Some report dramatic productivity gains (self-estimated up to 25×) and claim “software is dead” for typical CRUD/API/UI work; others suspect overestimation and marketing/astroturfing around “coding from your phone.”